def _build_plan(self, plan_factory, device_params, alpha, beta, seed):
plan = plan_factory()
bijection = philox(64, 2)
# Keeping the kernel the same so it can be cached.
# The seed will be passed as the computation parameter instead.
keygen = KeyGenerator.create(bijection, seed=numpy.int32(0))
sampler = normal_bm(bijection, numpy.float64)
squeezing = plan.persistent_array(self._system.squeezing)
decoherence = plan.persistent_array(self._system.decoherence)
plan.kernel_call(TEMPLATE.get_def("generate_input_state"),
[alpha, beta, squeezing, decoherence, seed],
kernel_name="generate",
global_size=alpha.shape,
render_kwds=dict(
system=self._system,
representation=self._representation,
Representation=Representation,
bijection=bijection,
keygen=keygen,
sampler=sampler,
ordering=ordering,
exp=functions.exp(numpy.float64),
mul_cr=functions.mul(numpy.complex128,
numpy.float64),
add_cc=functions.add(numpy.complex128,
numpy.complex128),
))
return plan
def add_param(arr_t, param_dtype):
"""
Returns an addition transformation with a dynamic parameter (1 output, 1 input, 1 scalar):
``output = input + param``.
"""
return Transformation(
[Parameter('output', Annotation(arr_t, 'o')),
Parameter('input', Annotation(arr_t, 'i')),
Parameter('param', Annotation(param_dtype))],
"${output.store_same}(${add}(${input.load_same}, ${param}));",
render_kwds=dict(add=functions.add(arr_t.dtype, param_dtype, out_dtype=arr_t.dtype)))
def add_param(arr_t, param_dtype):
"""
Returns an addition transformation with a dynamic parameter (1 output, 1 input, 1 scalar):
``output = input + param``.
"""
return Transformation(
[Parameter('output', Annotation(arr_t, 'o')),
Parameter('input', Annotation(arr_t, 'i')),
Parameter('param', Annotation(param_dtype))],
"${output.store_same}(${add}(${input.load_same}, ${param}));",
render_kwds=dict(add=functions.add(arr_t.dtype, param_dtype, out_dtype=arr_t.dtype)))
def add_const(arr_t, param):
"""
Returns an addition transformation with a fixed parameter (1 output, 1 input):
``output = input + param``.
"""
param_dtype = dtypes.detect_type(param)
return Transformation(
[Parameter('output', Annotation(arr_t, 'o')),
Parameter('input', Annotation(arr_t, 'i'))],
"${output.store_same}(${add}(${input.load_same}, ${param}));",
render_kwds=dict(
add=functions.add(arr_t.dtype, param_dtype, out_dtype=arr_t.dtype),
param=dtypes.c_constant(param, dtype=param_dtype)))
def add_const(arr_t, param):
"""
Returns an addition transformation with a fixed parameter (1 output, 1 input):
``output = input + param``.
"""
param_dtype = dtypes.detect_type(param)
return Transformation(
[Parameter('output', Annotation(arr_t, 'o')),
Parameter('input', Annotation(arr_t, 'i'))],
"${output.store_same}(${add}(${input.load_same}, ${param}));",
render_kwds=dict(
add=functions.add(arr_t.dtype, param_dtype, out_dtype=arr_t.dtype),
param=dtypes.c_constant(param, dtype=param_dtype)))
def test_multiarg_add(thr, out_code, in_codes):
"""
Checks multi-argument add() with a variety of data types.
"""
out_dtype, in_dtypes = generate_dtypes(out_code, in_codes)
def reference_add(*args):
res = sum(args)
if not dtypes.is_complex(out_dtype) and dtypes.is_complex(res.dtype):
res = res.real
return res.astype(out_dtype)
# Temporarily catching imaginary part truncation warnings
with catch_warnings():
filterwarnings("ignore", "", numpy.ComplexWarning)
mul = functions.add(*in_dtypes, out_dtype=out_dtype)
check_func(thr, mul, reference_add, out_dtype, in_dtypes)
def test_multiarg_add(thr, out_code, in_codes):
"""
Checks multi-argument add() with a variety of data types.
"""
out_dtype, in_dtypes = generate_dtypes(out_code, in_codes)
def reference_add(*args):
res = sum(args)
if not dtypes.is_complex(out_dtype) and dtypes.is_complex(res.dtype):
res = res.real
return res.astype(out_dtype)
# Temporarily catching imaginary part truncation warnings
with catch_warnings():
filterwarnings("ignore", "", numpy.ComplexWarning)
mul = functions.add(*in_dtypes, out_dtype=out_dtype)
check_func(thr, mul, reference_add, out_dtype, in_dtypes)
def _build_plan(self, plan_factory, device_params, alpha, beta, alpha_i,
beta_i, seed):
plan = plan_factory()
system = self._system
representation = self._representation
unitary = plan.persistent_array(self._system.unitary)
needs_noise_matrix = representation != Representation.POSITIVE_P and system.needs_noise_matrix(
)
mmul = MatrixMul(alpha, unitary, transposed_b=True)
if not needs_noise_matrix:
# TODO: this could be sped up for repr != POSITIVE_P,
# since in that case alpha == conj(beta), and we don't need to do two multuplications.
mmul_beta = MatrixMul(beta, unitary, transposed_b=True)
trf_conj = self._make_trf_conj()
mmul_beta.parameter.matrix_b.connect(trf_conj,
trf_conj.output,
matrix_b_p=trf_conj.input)
plan.computation_call(mmul, alpha, alpha_i, unitary)
plan.computation_call(mmul_beta, beta, beta_i, unitary)
else:
noise_matrix = system.noise_matrix()
noise_matrix_dev = plan.persistent_array(noise_matrix)
# If we're here, it's not positive-P, and alpha == conj(beta).
# This means we can just calculate alpha, and then build beta from it.
w = plan.temp_array_like(alpha)
temp_alpha = plan.temp_array_like(alpha)
plan.computation_call(mmul, temp_alpha, alpha_i, unitary)
bijection = philox(64, 2)
# Keeping the kernel the same so it can be cached.
# The seed will be passed as the computation parameter instead.
keygen = KeyGenerator.create(bijection, seed=numpy.int32(0))
sampler = normal_bm(bijection, numpy.float64)
plan.kernel_call(TEMPLATE.get_def("generate_apply_matrix_noise"),
[w, seed],
kernel_name="generate_apply_matrix_noise",
global_size=alpha.shape,
render_kwds=dict(
bijection=bijection,
keygen=keygen,
sampler=sampler,
mul_cr=functions.mul(numpy.complex128,
numpy.float64),
add_cc=functions.add(numpy.complex128,
numpy.complex128),
))
noise = plan.temp_array_like(alpha)
plan.computation_call(mmul, noise, w, noise_matrix_dev)
plan.kernel_call(TEMPLATE.get_def("add_noise"),
[alpha, beta, temp_alpha, noise],
kernel_name="add_noise",
global_size=alpha.shape,
render_kwds=dict(
add=functions.add(numpy.complex128,
numpy.complex128),
conj=functions.conj(numpy.complex128)))
return plan
def transformed_add(perf_params):
return functions.add(transformed_dtype(), transformed_dtype())
def _build_plan(self, plan_factory, device_params, output, alpha, beta):
plan = plan_factory()
samples, modes = alpha.shape
for_reduction = Type(alpha.dtype, (samples, self._max_total_clicks + 1))
prepared_state = plan.temp_array_like(alpha)
plan.kernel_call(
TEMPLATE.get_def("compound_click_probability_prepare"),
[prepared_state, alpha, beta],
kernel_name="compound_click_probability_prepare",
global_size=alpha.shape,
render_kwds=dict(
mul_cc=functions.mul(alpha.dtype, alpha.dtype),
exp_c=functions.exp(alpha.dtype),
))
# Block size is limited by the amount of available local memory.
# In some OpenCL implementations the number reported cannot actually be fully used
# (because it's used by kernel arguments), so we're padding it a little.
local_mem_size = device_params.local_mem_size
max_elems = (local_mem_size - 256) // alpha.dtype.itemsize
block_size = 2**helpers.log2(max_elems)
# No reason to have block size larger than the number of modes
block_size = min(block_size, helpers.bounding_power_of_2(modes))
products_gsize = (samples, helpers.min_blocks(self._max_total_clicks + 1, block_size) * block_size)
products = plan.temp_array_like(for_reduction)
read_size = min(block_size, device_params.max_work_group_size)
while read_size > 1:
full_steps = modes // block_size
remainder_size = modes % block_size
try:
plan.kernel_call(
TEMPLATE.get_def("compound_click_probability_aggregate"),
[products, prepared_state],
kernel_name="compound_click_probability_aggregate",
global_size=products_gsize,
local_size=(1, read_size,),
render_kwds=dict(
block_size=block_size,
read_size=read_size,
full_steps=full_steps,
remainder_size=remainder_size,
output_size=self._max_total_clicks + 1,
mul_cc=functions.mul(alpha.dtype, alpha.dtype),
add_cc=functions.add(alpha.dtype, alpha.dtype),
polar_unit=functions.polar_unit(dtypes.real_for(alpha.dtype)),
modes=self._system.modes,
max_total_clicks=self._max_total_clicks,
))
except OutOfResourcesError:
read_size //= 2
break
reduction = Reduce(for_reduction, predicate_sum(alpha.dtype), axes=(0,))
temp = plan.temp_array_like(reduction.parameter.output)
plan.computation_call(reduction, temp, products)
fft = FFT(temp)
real_trf = Transformation([
Parameter('output', Annotation(output, 'o')),
Parameter('input', Annotation(temp, 'i')),
],
"""
${input.ctype} val = ${input.load_same};
${output.store_same}(val.x);
""")
fft.parameter.output.connect(real_trf, real_trf.input, output_p=real_trf.output)
plan.computation_call(fft, output, temp, True)
return plan
